Method for establishing system resource prediction and resource management model through multi-layer correlations

Abstract

A method for establishing system resource prediction and resource management model through multi-layer correlations is provided. The method builds an estimation model by analyzing the relationship between a main application workload, resource usage of the main application, and resource usage of sub-application resources and prepares in advance the specific resources to meet future requirements. This multi-layer analysis, prediction, and management method is different from the prior arts, which only focus on single-level estimation and resource deployment. The present invention can utilize more interactive relationships at different layers to effectively perform predictions, thereby achieving the advantage of reducing hidden resource management costs when operating application services.

What is claimed is: 
     
       1. A method comprising:
 a) deploying and installing a main application on a plurality of nodes; 
 b) collecting, at predetermined intervals, workloads of the main application and usage amounts of every resource in the plurality of nodes used by the main application and a plurality of sub-applications thereof, and synchronously calculating first correlation values of a respective amount of each resource used by the main application to execute the workloads of the main application and second correlation values of a respective amount of each resource used by each sub-application to execute the workloads; 
 c) predicting the workload of the main application at time point T+1 in the future by a time series model at time point T, and determining the resource with a corresponding first correlation value higher than a first threshold value; and 
 d) creating a usage amount predictive model to predict usage amounts of every resource of the main application at each time point after time point T with usage amounts of every resource of the main application collected before time point T and inputting the predicted workload of the main application at time point T+1 into the usage amount predictive model to obtain a predicted increment of the usage amount of the resource of the main application found in the previous step at time point T+1, 
 wherein a calculating method of the first correlation value is calculating similarity measurement values with collected usage amounts of every resource of the main application and workloads of the main application, wherein when the similarity measurement value is negative, taking an absolute value thereof, wherein the similarity measurement value is obtained by calculating a cosine value from vectors formed by two change values of usage amounts of one single resource of the main application in three successive collecting time points and formed by two change values of workloads of the main application in three successive collecting time points. 
 
     
     
       2. The method according to  claim 1 , further comprising a step after the step d): e) assigning at least one additional node for the main application to use at time point T+1. 
     
     
       3. The method according to  claim 2 , further comprising a step of c1) after the step c) and a step of e1) after the step e):
 c1) determining corresponding sub-applications and sub-application related resources with a corresponding second correlation value higher than a second threshold value, and calculating ratios of said sub-applications relative to the main application using the sub-application related resource; and 
 e1) assigning usable amounts of the sub-application related resource in the at least one node to the corresponding sub-application according to the corresponding ratio at time point T+1. 
 
     
     
       4. The method according to  claim 3 , wherein the second threshold value is 0.5. 
     
     
       5. The method according to  claim 3 , wherein a calculating method of the second correlation value comprises the steps of:
 calculating a similarity measurement value between the usage amount of one single resource the main application and that of any sub-application, a similarity measurement value between usage amounts of any two resources of the main application, and a similarity measurement value between usage amounts of any two sub-application resources with collected usage amounts of every resource of the main application and usage amounts of every resource of the sub-applications, wherein if the similarity measurement values are negative, taking an absolute value thereof, and wherein the similarity measurement value between usage amounts of a single resource of the main application or that of sub-application and itself is set to be 1; and 
 averaging the similarity measurement values of the usage amounts of the single resource of the main application or that of the single sub-application. 
 
     
     
       6. The method according to  claim 5 , wherein the similarity measurement value is obtained by calculating a cosine value from vectors formed by two change values of usage amounts of the single resource of the main application in three successive collecting time points and formed by two change values of usage amounts of the single resource of any application in three successive collecting time points, vectors formed by two change values of usage amounts of any two resources of the main application in three successive collecting time points, or vectors formed by two change values of usage amounts of the single resource of any two applications in three successive collecting time points. 
     
     
       7. The method according to  claim 3 , further comprising a step of d1) after the step d) and a step of e2) after the step e1):
 d1) calculating an importance of weight for each sub-application regarding each sub-application related resource at time point T+1; and 
 e2) if the assigned usable amount of the sub-application related resource in the at least one node at time point T+1 is not able to meet the need for the related sub-application, prioritize the need for the sub-application related resource for the sub-application with a larger importance of weight. 
 
     
     
       8. The method according to  claim 7 , wherein the importance of weight is an average value of the second correlation value of any sub-application and the ratio of the sub-application relative to the main application in using the sub-application related resource at time point T. 
     
     
       9. The method according to  claim 1 , wherein the first threshold value is 0.5. 
     
     
       10. The method according to  claim 1 , wherein the usage amount predictive model uses a machine-learning algorithm to analyze usage amounts of every resource of the main application collected before time point T to predict usage amounts of every resource of the main application at every time point after time point T. 
     
     
       11. The method according to  claim 10 , wherein the machine learning algorithm is a Regression Analysis algorithm, Bayesian Belief Network algorithm, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, Q-learning algorithm or Poly Regression algorithm. 
     
     
       12. The method according to  claim 1 , wherein the time series model is an Autoregressive Integrated Moving Average (ARIMA) model. 
     
     
       13. The method according to  claim 1 , wherein the resource is a number of usable CPU cores, usable memory, usable storage, or usable network bandwidth.